Attribution window

An attribution window is the period after a change during which an observed effect is credited to that change.

Example

You update a product page on the first of the month. If an AI engine takes several days to re-crawl the page and reflect it in answers, a measurement taken the next day would miss the effect. A window that starts a few days later and runs for several weeks gives the change a fair chance to show up.

Why it matters

Start the window too early and you measure a period when the engine could not yet have seen the change. Make it too long and other events, such as competitor updates or model changes, mix into the result. AI engines can differ in how quickly they reflect new content, so one window for every engine can be too short for some and too long for others.

Attribution windows work best alongside a holdout control, which accounts for changes during the window that have nothing to do with your fix.

How Stellarcast handles it

Stellarcast opens a separate attribution window for each engine after a fix is published, starting after an engine-specific delay that allows time to re-crawl and re-answer. A result needs at least 14 days in the window, with enough actual readings, before it can be labeled MEASURED; a result that does not clear that bar is labeled LIKELY or TOO-EARLY.

Related terms

• Holdout control
• AI citation decay
• P25-P75 range
• AI referral traffic
• How Stellarcast measures AI visibility